AI Chatbot for Financial Advisory at maiwerk

Starting Point

maiwerk is an independent financial consultancy based in Cologne. Their advisory is transparent, independent and deeply human – not product sales, but real guidance on decisions that shape their clients' lives. That kind of advisory takes time, attention and an appointment.

And that is exactly where the gap opens up: between sessions, questions come up that should be answered quickly – and potential clients are looking for orientation outside office hours. Anyone who wants to make financial advisory more accessible without diluting its quality and character needs a way to make the existing advisory knowledge usable – without replacing the human part.

Delivery

Together with the maiwerk team, we are building an AI chatbot that sits on top of maiwerk's structured knowledge base. Through Retrieval-Augmented Generation (RAG), the bot draws on the editorial content and produces answers that don't sound generic but carry the tone and values of maiwerk's advisory. Delivery runs iteratively: every conversation becomes the basis for the next refinement.

• RAG pipeline on maiwerk's knowledge base – structured advisory content as the source, not generic web answers
• Context Design as the quality lever – iterative refinement of answer templates, so responses are not just correct, but relevant, personal and genuinely helpful
• Clearly scoped responsibilities – the bot knows when it can answer and when a human advisor is needed
• Modular content maintenance – maiwerk can add or update content without engineering involvement
• Continuous evaluation – every conversation yields signals about where answer quality is off, feeding back into Context Design

Technologies in Use

The platform combines an LLM-based conversation engine with a RAG pipeline on maiwerk's knowledge base. Embeddings are managed in a Vector Store that enables fast, precise semantic search across the advisory content. Context Design – the deliberate shaping of prompt and response patterns – is the quality lever that turns a generic LLM into an advisory tool aligned with maiwerk's standards. Conversation evaluation is part of the pipeline, ensuring answer quality does not stagnate but grows systematically over time.

Results

The project is an ongoing collaboration – impact shows up along the entire client journey:

• Scalable expertise – advisory quality is no longer bound by free slots in the calendar
• Greater accessibility – financial literacy is available when and where clients need it
• Trust through consistency – answers reflect maiwerk's values and language, not the generic tone of any random model
• Intelligent handovers – the bot knows its limits and hands off clearly to human advisory where required

The decisive lever wasn't the model, it was the content: maiwerk has a structured, user-focused knowledge base – and that editorial discipline is what separates a mediocre chatbot from one that responds with precision.

Lessons Learned

First: the quality of a RAG system is not decided by the model choice but by the knowledge base. Structured, clearly written advisory content beats any prompt engineering.

Second: Context Design is not "done after setup." Every conversation reveals where answer patterns are still too generic, too cautious or too technical. Anyone who doesn't run this loop systematically leaves the biggest quality lever on the table.